[virangjhaveri]

Do you reward the RL model based on the token consumption when multiple LLMs complete the task ?

[adchurch]

Effectively yes (based on cost though, not raw token count)